##########################################
Torch7 provides a Matlab-like environment for state-of-the-art machine
learning algorithms. It is easy to use and provides a very efficient
implementation, thanks to an easy and fast scripting language (Lua) and a
underlying C implementation.
torchandroid.h and torchandroid.cpp give lots of helper functions to make life easier
Print function overriden to redirect to logcat (only handles strings for now)
Function to get apk assets as bytes (very useful)
Full support for ffi and shared libraries
torch.load now takes three additional modes: apkbinary32, apkbinary64, apkascii. One can store model files in the assets folder and use these modes to load them. If the model was saved on a 64-bit machine, use apkbinary64, if it was saved on a 32-bit machine, use apkbinary32.
Extract it under ~/NVPACK, next to 12b that comes with CodeWorks.
Change NVPACK environvent to point to that NDK (see sample in ./.bashrc-android)
Samples
Three sample projects has been provided in demos/
demos/android-demo/jni/torchdemo.cpp is a simple use-case
demos/android-demo/assets/main.lua is the file that is run
demos/android-demo-cifar showcases classifying Camera inputs (or images from gallery) into one of 10 CIFAR-10 categories.
Vinayak Ghokale from e-lab Purdue (https://github.com/e-lab) contributed a face detector demo, which showcases a fuller use-case (demos/facedetector_e-lab ).
Building Torch Libraries and Java class.
If on ubuntu, install the following packages: sudo apt-get install libx32gcc-4.8-dev libc6-dev-i386Default is to build with CUDA - so make sure you installed NVIDIA CodeWorks for Android and its nvcc is in your PATH.
Otherwise, set WITH_CUDA=OFF in build.sh
git submodule update --init --recursive
Optionally, open build.sh and modify ARCH (to match your device architecture) and WITH_CUDA variables.
run build script:
3 ./build.sh
You can use torch in your android apps. The relevant directories are
install/include - include directories
install/libs/$APP_ABI - static libs cross-compiled for your APP_ABI
install/share/lua - lua files
Building Android Demo App
Build Torch-Android atleast once using the steps above.
[Optional] Connect your android phone in debugging mode,
to automatically install the apk.